Sentiment Analysis of Code-Mixed language

Abstract

Code-mixed language is very commonly used in today’s multilingual society. It is the phenomenon of mixing the syntax and vocabulary of many languages in single sentence. Sentiment analysis of code-mixed language aims at identifying the polarity value of the sentence. This paper mainly focuses on sentiment analysis of Tweets consisting of words from Hindi and English language along with other symbols. The dataset contains 20,000 tweets. We generate word level, character level and subword level representation for the Tweets which are used as input to the different models such as CNN, LSTM and BiLSTM. The performance of BiLSTM model is better as compared with other models. The accuracy of WORD Level BiLSTM, BPE Level BiLSTM and WORD Level CHAR Level BiLSTM is 60.27%, 58.59% and 54.24% respectively.